One of Netflix's many accomplishments is really driving personalized recommendations for its viewers. Given the growing breadth of digital libraries, navigation and discovery are becoming substantial challenges for the average viewer simply browsing for something appealing.
The company actively analyzes practically every click by every one of its 100 million-plus users to get a unique level of granular insight into viewing behavior and preferences. It's been aggressive in mining that data to create a growing library of original hit shows and a user experience that has been praised by most.
According to a recent report in The Register, the OTT giant is experimenting with ways to take this further. It is using machine learning and artificial intelligence (AI) to create personalized trailers for its shows and movies. The idea is to find the most compelling scenes in a TV show for a user and put them together, to make the most convincing trailer possible for that individual viewer. The example offered for Netflix's direction is that of an action movie that reaches out to "chick-flick" viewers by featuring a romantic scene in the trailer.
Apparently, a similar effort using AI to create a trailer has been done before -- by IBM for horror movie Morgan. However, in this case, it was about using AI to streamline the trailer-making process. IBM still used human editorial input. But with Netflix's scale, it will probably need the process to be almost, if not entirely, automated.
This report also follows a recent announcement from Telefonica regarding the integration of its AI platform Aura with its pay-TV service. In Telefonica's case, the goal was personalized advertising as well as enabling e-commerce and video sharing through Aura, which integrates the operator's various network systems, back-office and other IT systems, and delivery infrastructure. (See Telefónica Creates AI Aura Around Pay-TV.)
I would imagine the immediate goal for Netflix would be to develop a set of user profiles based on the titles viewed, and then categorize scenes in each title based on elements that would match up to the attributes included in each profile. Netflix can then select the right clip, with the right elements to match some combination of user preferences, and these can be pulled into the trailer.
UK Broadcaster Channel Four has developed a similar approach for its All 4 online service. According to the head of the service, Sarah Milton, the team has developed a segmentation scheme based on nine different kinds of viewers based on and supported by the algorithms. This caters to the main differences between the major viewer types while still a manageable number. But in All 4's case, the algorithm is used to recommend titles, and is only partly fed by data. The broadcaster also uses human editors to shape the selection. (See Channel Four's Milton Talks Personalization.)
Using AI and machine learning for search, discovery and navigation is an obvious use of the technology. Video-on-demand and the integration of OTT services on smart TVs and pay-TV services have made selection a far more complex exercise for viewers so it's important to help them find the content they want. Improvements in content discovery will help drive usage and appreciation of the service, which in turn will cut churn.
The technology sounds extraordinarily exciting, but I do wonder about this particular use of it, i.e., personalized trailers. That's because the movie itself will still be the same. So if you are the person in the example above, looking for a romantic film, and decide to watch a featured movie because the trailer features a romantic scene, where does that leave you? The movie doesn’t change; it’s still going to be an action film. What happens after eight car chases, 23 explosions and 18 straight minutes of gunfire? Most likely, you will have a headache and will never trust Netflix's recommendations again.
Not seeing the intelligence in that, artificial or otherwise.
— Aditya Kishore, Practice Leader, Video Transformation, Telco Transformation